RetroUSB AVS - HDMI NES implemented on a FPGA!

[Einzelherz]

That door hinge gon' break...

[Shoryukev]

This looks like a great option for people not wanting to use emulation and get good results on an HDTV, without buying an expensive pre-modded system or going through the trouble of modding it themselves. I'd be interested to check one out once it's released, the price-point seems pretty good

I will stick with my NESRGB, but I can see myself recommending this to several of my friends if it turns out to be a good product.

That door hinge gon' break...

I know it's probably an aesthetic choice making the door long enough so the entire cartridge label is visible when it's open, but if someone were to lift up on the cart when it's inserted I'd be worried about the 72-pin connector.

[Einzelherz]

This looks like a great option for people not wanting to use emulation and get good results on an HDTV, without buying an expensive pre-modded system or going through the trouble of modding it themselves. I'd be interested to check one out once it's released, the price-point seems pretty good

I will stick with my NESRGB, but I can see myself recommending this to several of my friends if it turns out to be a good product.

That door hinge gon' break...

I know it's probably an aesthetic choice making the door long enough so the entire cartridge label is visible when it's open, but if someone were to lift up on the cart when it's inserted I'd be worried about the 72-pin connector.

Yup, I saw that, too. It's almost like having a small door at the front was for a reason!

It's a decent enough idea, although the 4 player option seems a bit silly, but the design has always irked me.

[tacoboy42]

I know it's probably an aesthetic choice making the door long enough so the entire cartridge label is visible when it's open, but if someone were to lift up on the cart when it's inserted I'd be worried about the 72-pin connector.

it was designed that way to be able to fit the famicom disk RAM adapter, and the cart sticks in enough that would prevent someone from lifting straight up, it would need to come out before being able to lift it up.

[lettuce]

This looks like a great option for people not wanting to use emulation and get good results on an HDTV, without buying an expensive pre-modded system or going through the trouble of modding it themselves.

Its not using emulation though!

[Shoryukev]

This looks like a great option for people not wanting to use emulation and get good results on an HDTV, without buying an expensive pre-modded system or going through the trouble of modding it themselves.

Its not using emulation though!

Hence why I said it looks like a great option for people that don't want to use emulation.

[Guspaz]

I think you mean it is emulation? Hardware emulation is still emulation, and there's still the question of how accurate the emulation is. Then again, consider that the NESRGB and Hi-Def NES both essentially replace parts of the NES with emulated counterparts, particularly in the case of the Hi-Def NES where the entire audio system is emulated.

[Shoryukev]

I think you mean it is emulation? Hardware emulation is still emulation, and there's still the question of how accurate the emulation is. Then again, consider that the NESRGB and Hi-Def NES both essentially replace parts of the NES with emulated counterparts, particularly in the case of the Hi-Def NES where the entire audio system is emulated.

Is clone hardware considered emulation? I've never really thought of it like that before, but the way the NESRGB works alongside the original PPU could probably be considered a form of emulation as well

[bobrocks95]

Of course the alternatives for people who want that are the Hi-Def NES which is going to be more expensive once you add in installation cost, the price of a NES console, a four score and a cartridge adapter. Potentially less durable too since you would be reusing old parts? You also have NESRGB, but then you need a decent video scaler (200$-ish) and still the additional costs above.

People who want none of the audio/video latency overhead of a OS running an emulator. This thing doesn't really compare to the Retron.

Of course, but I think it's fair to assume that those people who don't like emulators' flaws already own an NES. In which case adding the Hi-Def NES is much closer in cost to getting this. I think the missing link I was neglecting was maybe native Famicom support. So then it seems like it's limited to those who want to buy Famicom stuff and/or don't want to deal with the hassle of installing or getting someone to install the Hi-Def NES.

Maybe I'm underestimating the amount of people who don't want to mod their consoles, or the number of people who wouldn't mind buying a whole new console when they already have an original?

[Guspaz]

I think you mean it is emulation? Hardware emulation is still emulation, and there's still the question of how accurate the emulation is. Then again, consider that the NESRGB and Hi-Def NES both essentially replace parts of the NES with emulated counterparts, particularly in the case of the Hi-Def NES where the entire audio system is emulated.

Is clone hardware considered emulation? I've never really thought of it like that before, but the way the NESRGB works alongside the original PPU could probably be considered a form of emulation as well

Sure: neither clone hardware nor an FPGA implementation work the same way internally as the original hardware, they're attempting to re-implement the original hardware such that the same inputs produce the same outputs in the same amount of time. That's also a description of software emulation.

[RGB32E]

Sure: neither clone hardware nor an FPGA implementation work the same way internally as the original hardware, they're attempting to re-implement the original hardware such that the same inputs produce the same outputs in the same amount of time. That's also a description of software emulation.

And you speak from what working experience with FPGA development?

[Guspaz]

And you speak from what working experience with software development? An FPGA is physically different from the original hardware, it is a simulacrum pretending to be the same, as in the Chinese room thought experiment. It remains emulation, regardless of whatever trolling or ad hominem attacks you engage in.

[RGB32E]

Well, you can get an integer scale to 4K from 720p, but so few display manufacturers do this it's kind of a moot point.
I think the menu is the real killer here personally speaking, but not as universal a problem.

Do tell us some model numbers of such displays! Sure, the higher end Panasonic UHDTVs have a 1080p_by_4pixel mode for 1080p 2x integer scale, but I've seen no indication that this applies to 720p sources. On top of that, these models are no longer sold in the US and I've yet to see one with input lag under 50ms!

On a related topic, it was both entertaining and disappointing to learn that the XBoneS 2160p upscaling is mediocre (surprise, surprise).
http://www.eurogamer.net/articles/digit ... e-s-review

In the here and now, the upscaling from 1080p to 4K is competent but could be better. Hooking up the Xbox One S to a 58-inch Panasonic DX750, we noted that the UI upscaled to 4K better using the display's scaler compared to the Xbox One S's efforts (yes, the front-end isn't native UHD).

We would really like to see a 4x4 scaler mode added - it's effectively a simple 'nearest neighbour' upscale but it works well because 4K is a simple 2x scale in both directions compared to full HD. The feature is supported on our Panasonic DX750, and at range, it's definitely better for gaming than both the Xbox One S and indeed the DX750's own standard upscaler.

It's an awesome product, just wondering what role it fills. Thoughts?

Fans of RetroUSB who want a Retron 5 alternative?

[RGB32E]

Guspaz, you're just posting foolish speculation and I'm calling you on it!

[Guspaz]

Definitions are not speculation, why would you think that an FPGA emulating a chip was anything but emulation?

You may find this article on hardware emulation interesting: https://en.wikipedia.org/wiki/Hardware_emulation

EDIT: I should clarify, I'm not saying that software and hardware emulation work the same way, only that they are both simulations of the original hardware, and are only as accurate as their implementation. And the accuracy of the AVS' implementation is not something that I'm speculating on, I've no idea how accurate it is.

[Unseen]

Sure: neither clone hardware nor an FPGA implementation work the same way internally as the original hardware

Hypothetical questions: What if the FPGA implementation would be based on the "inner workings" of the original? For example, an FPGA version of the 6502 based on the data gathered by the Visual6502 project (de-capped original chip, node list and layout reversed from that) or if someone would take the original Pong schematics and put it on an FPGA either by describing the exact circuit in VHDL/Verilog or using Schematic-to-FPGA tools?

Or the other way around, is a revised original hardware that uses a slightly different implementation (e.g. 1-Chip-SNES or later versions of the PS1 that changed from SGRAM to SDRAM) an emulation?

they're attempting to re-implement the original hardware such that the same inputs produce the same outputs in the same amount of time. That's also a description of software emulation.

I would argue that there is still an important difference: Software emulation needs to virtualize time, because it has to execute the emulator on a CPU that is based around sequential program execution while the hardware it emulates can do multiple things at the same time. The software-based emulator only appears to work in real time because it occasionally waits to ensure that the time of the virtual system matches up with the real time and this only needs to happen on a "human-perceptible scale". In a hardware implementation, time must be explicitly managed, so the emulation can match the original hardware not just on a functional but also temporal level, with single clock cycle accuracy if such care is taken in the hardware implementation.

[bobrocks95]

Well, you can get an integer scale to 4K from 720p, but so few display manufacturers do this it's kind of a moot point.
I think the menu is the real killer here personally speaking, but not as universal a problem.

Do tell us some model numbers of such displays! Sure, the higher end Panasonic UHDTVs have a 1080p_by_4pixel mode for 1080p 2x integer scale, but I've seen no indication that this applies to 720p sources. On top of that, these models are no longer sold in the US and I've yet to see one with input lag under 50ms!

I was just saying it's entirely possible to do simple nearest-neighbor scaling. I have no interest in a 4K TV right now so I don't follow their developments- surely a 4K PC monitor from some company can do an integer scale from 720p?

[Guspaz]

Sure: neither clone hardware nor an FPGA implementation work the same way internally as the original hardware

Hypothetical questions: What if the FPGA implementation would be based on the "inner workings" of the original? For example, an FPGA version of the 6502 based on the data gathered by the Visual6502 project (de-capped original chip, node list and layout reversed from that) or if someone would take the original Pong schematics and put it on an FPGA either by describing the exact circuit in VHDL/Verilog or using Schematic-to-FPGA tools?

Sure, because you've still got a bunch of little programmable logic blocks pretending to be the original hard-wired implementation. It's still emulation, but if it's perfectly cycle-accurate and running at the same clockspeed, does it really matter?

Or the other way around, is a revised original hardware that uses a slightly different implementation (e.g. 1-Chip-SNES or later versions of the PS1 that changed from SGRAM to SDRAM) an emulation?

In a sense, yes, and the 1-chip is also of questionable accuracy considering that there are glitches in various games. That's getting into an argument of authenticity, though, rather than definition. I would argue that the 1-chip is authentic even if it isn't accurate.

they're attempting to re-implement the original hardware such that the same inputs produce the same outputs in the same amount of time. That's also a description of software emulation.

I would argue that there is still an important difference: Software emulation needs to virtualize time, because it has to execute the emulator on a CPU that is based around sequential program execution while the hardware it emulates can do multiple things at the same time. The software-based emulator only appears to work in real time because it occasionally waits to ensure that the time of the virtual system matches up with the real time and this only needs to happen on a "human-perceptible scale". In a hardware implementation, time must be explicitly managed, so the emulation can match the original hardware not just on a functional but also temporal level, with single clock cycle accuracy if such care is taken in the hardware implementation.

That's an implementation detail, however, and the key section is "if such care is taken". Accuracy is not guaranteed with hardware emulation, but this is not to say that perfectly accurate emulation is not possible.

I'd also point out that a sufficiently fast CPU running a baremetal software emulator that was designed for that specific host CPU can also achieve cycle-level accuracy, since it knows how many cycles the instructions took on the original processor, and how many cycles it takes to emulate that instruction on the host processor. Of course, this is a contrived and unrealistic scenario.

[RGB32E]

I was just saying it's entirely possible to do simple nearest-neighbor scaling. I have no interest in a 4K TV right now so I don't follow their developments- surely a 4K PC monitor from some company can do an integer scale from 720p?

I've been following and unfortunately no manufacturer is providing such an option. Such an option would make the AVS a very viable option on a 4k TV!

a contrived and unrealistic scenario.

As expected from you!

[Guspaz]

And such trolling is of course expected from you.

[Arasoi]

Get a room, you two.

Er, anyway. I broke down and pre-ordered one of these, it looks very promising for an NES solution for my plasma screen in the living room. I have ~ 500 games for NES/FC/FDS so I'll be putting it through it's paces when it arrives.

Some of the features in it are pretty nice, and I personally like the design on it. I share the same fears others have about support down the line but I think it will work well for what it is in the long run.

[Pasky]

Guspaz is correct for the most part. He's providing a constructive argument rather than "NO YOU'RE WRONG AND I'LL LET OTHERS SAY WHY"

Another good question is are transistors the lowest possible level you need to begin emulation at? You can emulate actual transistors, as this is done in microprocessor R&D when they're trying to understand how to make the transistors or trying to debug a chip. There are weird analog electrical effects that make transistors behave differently than assumed, non-discrete timing effects that make things happen in slightly different sequences according to chance or temperature, or tiny manufacturing differences. So things like Visual6502 assumes things about time, space and electricity and makes transistors behave ideally. Maybe you're splitting hairs at that point... but that's emulation philosophy 101.

[ZellSF]

On 4K TVs: if the TV doesn't have an internal integer scale option then 1080p is just 1x more over 720p. A device with 1080p output won't look that much crisper on a 4K display than one with 720p output because they're both heavily interpolated anyway.

IMO integer scaling is heavily overrated anyway. As long as you have a 720p source it will look fantastic with a decent scaling algorithm.

Of course the alternatives for people who want that are the Hi-Def NES which is going to be more expensive once you add in installation cost, the price of a NES console, a four score and a cartridge adapter. Potentially less durable too since you would be reusing old parts? You also have NESRGB, but then you need a decent video scaler (200$-ish) and still the additional costs above.

People who want none of the audio/video latency overhead of a OS running an emulator. This thing doesn't really compare to the Retron.

Of course, but I think it's fair to assume that those people who don't like emulators' flaws already own an NES. In which case adding the Hi-Def NES is much closer in cost to getting this. I think the missing link I was neglecting was maybe native Famicom support. So then it seems like it's limited to those who want to buy Famicom stuff and/or don't want to deal with the hassle of installing or getting someone to install the Hi-Def NES.

Maybe I'm underestimating the amount of people who don't want to mod their consoles, or the number of people who wouldn't mind buying a whole new console when they already have an original?

Uh, the assumption that everyone who doesn't like emulators already have a NES is not a fair assumption, we have new people coming to this hobby all the time.

Also even if you have a NES already, I'm not so sure getting it modded is price competitive with the AVS. One option I see is 120$ for the hardware and 65$ for the mod process. That's already the price of the AVS, but you also have to pay for shipping your console (and your time doing that) to the modder. Plus add more money if you want to play Famicom cartridges, want your console recapped or want the unreliable cartridge connector replaced (and as mentioned earlier, if you want four controller ports).

If you're a DIY type that puts zero value on your own time then the Hi-Def NES might be a cheaper option, but if you're not? I'm not seeing that. Much more convenient and probably cheaper to buy a AVS and donate that old NES you have to someone who doesn't mind composite video quality.

That's still under the assumption that the AVS does what it promises. It might turn out not to be a good replacement for a NES, time will tell.

[Unseen]

Sure, because you've still got a bunch of little programmable logic blocks pretending to be the original hard-wired implementation. It's still emulation, but if it's perfectly cycle-accurate and running at the same clockspeed, does it really matter?

Interesting, I think that means that for you the "emulation or not" question might be based on the presence or absence of one or many switches. =)

Just for fun, let's take this to a few extreme cases: Let's assume we have a synthesizable implementation of some circuit in a hardware description language (HDL) like VHDL or Verilog. This can obviously be used to generate both a bitstream for a specific FPGA and also an ASIC that implements this circuit and according to your argument the FPGA would be an emulation while the ASIC is not. However, these are not the only options to put a circuit into silicon - for example Altera had a product line called "HardCopy" which used chips that were structured the same as an FPGA, but instead of loading a configuration bit stream into it, the function of the circuit was hardwired by implementing the configuration store as a ROM instead of RAM. Coming from the ASIC side, there are not just full-custom ASICs where you define everything from the transistor up, but also semi-custom ASICs which are designed from standard building blocks - some of them as small as a single gate, some as large as a complete microprocessor (e.g. Capcom's Kabuki security chip). There even have been synthesis tools that could output circuits based around arbitrary basic logic functions, so in theory we could create a circuit with discrete TTL logic chips from the HDL description. Someone also designed an FPGA-like logic block from 74xx logic for illustration purposes, but since that is also soft-configured we can just file this under FPGA for these purposes.

(and for really annoying nitpicking one could assume that the HDL-to-TTL implementation is built on solderless breadboards which are technically not hardwired, but technically don't have switches either ;) )

In a sense, yes, and the 1-chip is also of questionable accuracy considering that there are glitches in various games. That's getting into an argument of authenticity, though, rather than definition. I would argue that the 1-chip is authentic even if it isn't accurate.

There are some other examples: The C64 originally used a 6510 processor, but later models switched to an 8500. Originally it was assumed that this change in designation was just because the chip was moved to a newer fab process, but people who know more about this that I do claim that based on die shots the 6510 was a classic hand-layouted chip, while the 8500 was re-done in CAD with standard cells (i.e. gates of the same type are identical everywhere). As far as I know, nobody has yet found a way to distinguish between these two chips purely from software.

That's an implementation detail, however, and the key section is "if such care is taken". Accuracy is not guaranteed with hardware emulation, but this is not to say that perfectly accurate emulation is not possible.

That was just a disclaimer because there are some HDL implementations that use a similar "run-and-wait" approach as a software emulator does to provide cycle-exact emulation. A common tell-tale sign of this is that they need a mugh higher input clock than the original chip even after figuring in that the original may for example triggered its logic on both clock edges or internally generated multiple phase-shifted clock signals to divide a clock cycle into smaller parts.

I'd also point out that a sufficiently fast CPU running a baremetal software emulator that was designed for that specific host CPU can also achieve cycle-level accuracy, since it knows how many cycles the instructions took on the original processor, and how many cycles it takes to emulate that instruction on the host processor. Of course, this is a contrived and unrealistic scenario.

Maybe we should throw power efficiency into the argument too? ;) IMHO it's still a fundamenta difference: Although a processor can achieve the same if it is fast enough, it's still serial processing while the FPGA can achieve the same amount of parallelism as the original hardware.

Another good question is are transistors the lowest possible level you need to begin emulation at?

Transistors are already an abstraction, you could also use a field solver to calculate the flow of electrons within a 3D representation of the chip's silicon. It would probably a bit slow though. ;)

You can emulate actual transistors, as this is done in microprocessor R&D when they're trying to understand how to make the transistors or trying to debug a chip.

For debugging a chip, you would not go as far down as emulating actual transistors - even that is way too slow. I know someone who ran a few simulation of a simple adder circuit modelled from transistors in SPICE (an analog circuit simulator) to determine how susceptible the calculated results are to various condition like too-low voltage and that took quite a few hours on the University's compute cluster.

There are weird analog electrical effects that make transistors behave differently than assumed, non-discrete timing effects that make things happen in slightly different sequences according to chance or temperature, or tiny manufacturing differences. So things like Visual6502 assumes things about time, space and electricity and makes transistors behave ideally.

Indeed - Visual6502 uses what is called a "switch-level transistor model". The transistors are modelled as simple switches that are either on or off, depending on the logic level they see on their gate. It's already a very slow model, e.g. about 10kHz or so for the 6502 with a simulator written in C and although it passes a validation suite for all official opcodes, there are some differences in this model compared to a C64 for some of the "undocumented"/"illegal" opcodes. I'm not sure though if these differences are the result of inaccuracies in the switch-level modelling or due to the presence of a second bus master (the video chip) in the C64 - the simulated environment I used to run the test suite only had simulated RAM and the CPU model.

[RGB32E]

Until people can actually compare the AVS to original HW revisions, it's purely speculation to ramble on about the connotations of the word "emulation" in relation to FPGA development (naïve misnomer).

Get a room, you two.

I know, right? Guspaz and Pasky are still pretty salty!

Er, anyway. I broke down and pre-ordered one of these, it looks very promising for an NES solution for my plasma screen in the living room. I have ~ 500 games for NES/FC/FDS so I'll be putting it through it's paces when it arrives. Some of the features in it are pretty nice, and I personally like the design on it. I share the same fears others have about support down the line but I think it will work well for what it is in the long run.

Congrats! I'm looking to see how it performs as well! Brian seems committed to supporting the console. I'm very curious to listen to how expansion audio is handled given that the AVS has an ADC stage so that expansion audio doesn't have to be implemented on the FPGA like Kev did with the HDN.

[Arasoi]

.

[RGB32E]

Congrats! I'm looking to see how it performs as well! Brian seems committed to supporting the console. I'm very curious to listen to how expansion audio is handled given that the AVS has an ADC stage so that expansion audio doesn't have to be implemented on the FPGA like Kev did with the HDN.

I've been looking around, but can't find confirmation.. does the console support 480p out over HDMI? If it does then I don't think 240p would be off the table with a lagless active HDMI to VGA adapter and a Mimo Genius/UVC/Extron Super Emotia GX setup. Not it's purpose I understand, but a nice possible bonus for a handful of folks like me.

720p only. Brian's AVP (AVS portable) runs at 480p. I'd imagine he could add 480p output to the AVS if he wanted to, but guessing he won't.

[Guspaz]

Having experimented a bit with the different approaches to upscaling, I've come to the conclusion that 720p output (whereby an integer scale is used from 240p to 720p and some other filter such as bilinear is used for 720p to 1080p) is actually preferable to 1080p nearest-neighbour. It's the best of both worlds: more than sharp enough with distinct source pixels, but the slightly softer edges are easier on the eyes, even-sized pixels (instead of the 4/5/4 pattern of a 4.5x nn scale), more likely for scanlines to look right, and cheaper to implement.

I've seen a fair amount of complaints about the AVS not supporting 1080p, but in my mind, 1080p output is unimportant, and it isn't the resolution of my primary display anyhow (1440p).

[tusecsy]

Having experimented a bit with the different approaches to upscaling, I've come to the conclusion that 720p output (whereby an integer scale is used from 240p to 720p and some other filter such as bilinear is used for 720p to 1080p) is actually preferable to 1080p nearest-neighbour. It's the best of both worlds: more than sharp enough with distinct source pixels, but the slightly softer edges are easier on the eyes, even-sized pixels (instead of the 4/5/4 pattern of a 4.5x nn scale), more likely for scanlines to look right, and cheaper to implement.

I've seen a fair amount of complaints about the AVS not supporting 1080p, but in my mind, 1080p output is unimportant, and it isn't the resolution of my primary display anyhow (1440p).

I don't have much experience with this, when my TV upscales a 720p to 1080p image, will the pixels appear as small as they do when running a 240p image upscaled to 1080p through the framemeister? Or will it look like the 720p output of the framemeister? I greatly prefer the look of the 1080p output on the framemeister to the 720p output so this would be a huge deal for me AVS wise. Although the NES image is definitely the least pronounced difference between the two resolutions xrgb-wise.

As for the unit itself, I feel the design is a little wonky but the retrousb dude is really good people and he fully explained the reasonings for his design, all of which make sense. I will be buying this product and his portable unit when it's released as well. I feel it's a great price point, a great alternative to the utter shite retron, and a shining example of small business ingenuity and growth.

As for the FPGA, if it looks and quacks like a duck, it's a duck. No need to argue semantics is there? From my understanding FPGA is the best money can buy when it comes to hardware recreation outside of the actual original components. Components that are going to crap left and right keep in mind. The reason I'm buying this thing is because a brand new looking and acting AV/Famicom I recently tried to get modded had screwed up video chips. And this thing worked and looked cherry. These things are 25-30+ years old now, it's time for a new product and it looks like the AVS is it.

[Guspaz]

I don't have a Framemeister, and it does depend on the exact scaling algorithm used by your monitor, but I did a comparison between three scenarios on getting a 240p image (well, 224p since it was a SNES screenshot) to 1080p.

Here is a straight-up 4.5x bilinear scale, direct from 240p to 1080p:

http://i.imgur.com/lb6efSc.png

Here is a 4.5x nearest-neighbour scale, direct from 240p to 1080p (non-integer scale, uneven pixel sizes):

http://i.imgur.com/S2OyprZ.png

And here is a multi-pass scale, a 3x nearest-neighbour scale followed by a 1.5x bilinear scale, to simulate the AVS or OSSC at 720p being scaled to 1080p by a display:

http://i.imgur.com/Mc2ul1l.png

I prefer the final screenshot. Different displays may use different scaling algorithms, however, so it's possible that a display with a really offensive scaler might look particularly bad. My monitor appears to use bilinear, but some TVs use something that looks a lot more like an edge-directed filter similar to Photoshop's "preserve detail" resampling. I should prepare some examples with bicubic and edge-directed.